lnwjud-scheduled-continuation

Start a durable lnwjud goal with run_goal before the first mutation, checkpoint milestones with checkpoint_goal, and ensure one recurring hourly Native ChatGPT Scheduled Task exists via prepare_scheduled_continuation. The goal then resumes on each hourly wake without the user typing continue.

Each active goal gets exactly one Native ChatGPT Scheduled Task with occurrence=interval and intervalMinutes=60 in the current chat. Ordinary wakes never consume or replace it; the same task ID survives until the goal completes and the task is deleted or disabled.

No. The skill explicitly forbids lnwjud scheduler, Windows Task Scheduler, schtasks.exe, cron, shell timers, and browser automation as fallbacks. Only the Native ChatGPT Scheduled Task surface resolved from the current host registry is allowed.

The next hourly wake calls claim_scheduled_continuation, which detects the stale lease after a bounded 60-second heartbeat grace window and recovers the orphan in the same tick. A new worker reacquires the same goalKey with run_goal and resumes from the latest checkpoint.

A checkpoint is not a turn boundary; it only persists progress. The worker must continue useful work in the same turn and may yield only when the goal is terminal, a real external blocker exists, or confirmed continuation coverage allows a safe boundary.

Completion requires all blocking tasks terminal with inspected results, the recurring native task deleted or disabled with a recorded cancellation receipt, and finish_goal returning completionState=completed confirmed by get_goal. Reporting done while get_goal is active is forbidden.